search

Amshaker / unetr_plus_plus

[IEEE TMI-2024] UNETR++: Delving into Efficient and Accurate 3D Medical Image Segmentation
Apache License 2.0
340 stars 32 forks source link

BraTs Data #37

Closed smanman closed 1 year ago

smanman commented 1 year ago

BraTs dataset in trainer train has 387 datasets, val has 73 datasets together a total of 460, but a total of 484, the author provides preprocessed imagesTs dataset is 485 after no labels, if not run directly run_evaluation_tumor.sh, run separately predict_simple.py At this time, pass in the parameter i What should the imagesTs path dataset be? Can you give a specific catalog training set, validation set, and test set? Or can I put all the data except for the test set (i.e. the val file of the source trainer) back under the trainer?

Amshaker commented 1 year ago

Hi @6018203135 ,

Please refer to our paper for the full details. We mentioned that we divided BraTs with 80:5:15 ratios for training, validation, and testing. This means that we have 387 training samples, 24 samples for internal validation (hyper-parameter tuning, not provided in the pre-processed dataset), and 73 samples for testing (which you call validation). You can follow the same folder structure of the pre-processed dataset without change.

I hope it is clear now. Best regards, Abdelrahman.

smanman commented 1 year ago

@Amshaker Thank you veay much!